Anton

What is Anton?

MindsDB Anton is a business intelligence agent you can chat with in plain language. Its core purpose is to help you go from a natural-language question to an end-to-end analytical result—pulling and unifying data from sources, running analysis, surfacing insights, and producing dashboards.

Anton is designed to handle the work of an analyst “from the conversation.” For example, when you ask for prices, calculations, performance windows, and a dashboard, Anton can fetch live data, generate and run the required code, and present the results in a browser without a separate manual setup step.

Key Features

• Credential vault for secrets: Keeps secret values hidden from LLMs by storing/revealing only secret names during data connections.
• Isolated code execution (“show your work”): Runs generated code in a protected, reproducible environment and can provide an explainable scratchpad-style breakdown.
• Multi-layer memory & continuous learning: Maintains session-level context plus semantic and long-term business knowledge using human-readable memory systems.
• Connect to local data via a secure workflow: Use a /connect command to add data sources; Anton fetches schema and retrieves needed data after looking up credentials in the vault.
• Explainable-by-default scratchpad: You can ask Anton to dump its scratchpad, including notebook-style logs of code cells, outputs, and errors.

How to Use Anton

1. Install Anton from the repository using one of the provided commands:
   • macOS / Linux (CLI): curl -sSf https://raw.githubusercontent.com/mindsdb/anton/main/install.sh | sh && export PATH="$HOME/.local/bin:$PATH"
   • Windows (PowerShell): irm https://raw.githubusercontent.com/mindsdb/anton/main/install.ps1 | iex
   • The README states you can then run it by typing anton.
2. Start a conversation by asking a business question in natural language (e.g., requesting portfolio valuation, stock performance, and a dashboard).
3. Connect your own data when you need private datasets:
   • Run /connect and follow prompts to choose a data source.
   • Provide/select credentials by name so Anton can use the values stored in the local vault.
4. Ask for an explanation when you want to review what happened:
   • Request that Anton dump its scratchpad for a step-by-step notebook-style breakdown.

Use Cases

• Portfolio and performance dashboards: Provide a holdings list (tickers/amounts) and ask for current prices, total portfolio value, and 30-day performance, with a dashboard generated from the conversation.
• Correlation analysis on internal data: Connect to a company database or app data, then ask whether a business metric (e.g., discount) correlates with another outcome (e.g., review rating) over a defined time window.
• Dashboard creation from multiple sources: Ask a single question that requires combining live/public data with connected internal data, so Anton handles schema discovery and retrieval.
• Reviewing and debugging analysis steps: When the generated analysis isn’t clear, ask Anton to explain by dumping its scratchpad to see code cells, outputs, and errors.
• Project-scoped context and repeatable workspace runs: Run Anton in a directory with a workspace layout that stores scratchpad state, episodic memories, and local secrets; optionally add project context via .anton/anton.md.

FAQ

Is Anton only for public data? Anton can be used with public data, but the README notes that the “real power” comes from combining public data with your own connected data sources.

How does Anton access my credentials? Anton uses a credential vault workflow with /connect, where secret values remain hidden and Anton can access credentials by their secret names.

Can I see how Anton reached its results? Yes. Anton supports explainable-by-default behavior: you can ask it to dump its scratchpad to get a notebook-style breakdown including code cells, outputs, and errors.

What platforms are supported based on the install instructions? The README provides installation steps for macOS/Linux (CLI) and Windows (PowerShell), and mentions a macOS Desktop App for downloading.

Where does Anton store its workspace data? When run in a directory, the README describes a .anton/ workspace folder containing scratchpad state, episodic memory, and local secrets, along with related files such as .anton/anton.md (optional project context) and .anton/.env.

Alternatives

• General-purpose data analysis notebooks (e.g., Python/Jupyter): Instead of an agent-driven conversation that fetches data and generates code on the fly, you manually write and run analysis, which can be better for fully controlled workflows.
• BI dashboard tools that require pre-modeled data: Tools focused on building dashboards often rely on curated data models and setup steps; Anton’s workflow emphasizes starting from plain-language questions.
• SQL-based analytics and ETL pipelines: If your team prefers deterministic, reviewable transformations and prefers writing SQL/ETL jobs directly, this can reduce runtime code generation compared to an AI agent approach.
• Other AI assistants for data querying: Adjacent tools may also translate questions into queries, but Anton’s standout positioning in the README is its explainable scratchpad, credential-vault approach, and local workspace memory systems.